A cost reckoning for AI is finally happening. A recent convergence of references and stories about the high cost of running AI systems at enterprise scale caught my attention over the last week. Together, they provide increasing evidence that the adoption of AI may hit a speedbump some time this year. Agentic AI has become impressively powerful, but the commensurate costs are gradually causing some pause.
Cost Reckoning
In “There’s no such thing as free AI” Marketplace concluded that “AI consumption costs have been rising to levels that rival human labor costs” based on a review of a Goldman Sachs research survey. While the concerns may be short-term given Gartner predicts a 90% reduction in token costs for LLMs (large language models) between 2025 and 2030, lower costs for powerful technology can still drive consumption to budget-busting levels. Cheaper and more powerful AI unlocks more and more possibilities and those possibilities in turn drive more consumption. Thus, the percolating cost concerns are well-founded.
CNBC ran a segment on “over-inflated” demand for AI. The sensationalistic title oversold a story about unsustainable demand. Still, I took note of some important proof points on the cost reckoning for AI:
- Uber Technologies’ (UBER) CTO said the company’s AI coding tools have already burned through its full-year AI budget (April 16, 2026).
- Goldman Sachs (GS) research claims that companies are overrunning inference budgets by orders of magnitude; AI costs will rival engineering headcount in 2026. {Marketplace referenced the same study}.
- “Tokenmaxxing” is creating perverse incentives to needlessly consume tokens to demonstrate AI adoption. {This was CNBC’s direct reference to over-inflated demand}.
- Anthropic removed OpenClaw from flat-fee subscriptions in Claude. A single $200 Anthropic plan can burn from $2000 to $5000 in compute. Thus, Anthropic is moving enterprise customers to token-based plans and away from seat-based plans.
CNBC anchor and columnist Deirdre Bosa concluded that “the companies that price for real demand are going to look different from the ones that didn’t.” In other words, costs are going up and everyone in the ecosystem will start to bear the true costs of AI consumption.
Bring Back the Humans
These costs undermine the most grandiose predictions of a future with no human employees. A future here AI writes all the software…and presumably AI Agents also buy and use all this software. Today’s companies in certain industries that are rushing to replace humans en masse with AI or even dreaming about such a move may soon run into the harsh reality of a cost reckoning.
Conclusion
I have increasingly wondered what will happen to AI consumption once consumers face the real costs of using AI-based systems. At some point, venture capital will stop subsidizing the consumption and distribution of tokens that are the lifeblood of AI systems. The cost reckoning awaits and the providers and users of AI may experience business shocks if they are not preparing now.
Of course, the frenzied rush to AGI (Artificial Generative Intelligence) will motivate the biggest spenders in the industry to keep spending. Venture capital is also chasing AI innovations in a gold rush moment for tech spending. According to Visual Capitalist, AI’s share of venture capital as quickly gone from 20-30% in 2022 and 2023 to over 50% in 2025.
Speculative spending may be enough to keep the boom going even as companies start to address the cost reckoning in AI. The fresh all-time highs in my favorite AI bundle, the Robo Global Artificial Intelligence ETF (THNQ), suggests the frenzy just accelerated. I continue to hold THNQ as my core AI investment.
Be careful out there!
Full disclosure: long THNQ